Motius Evaluator Humanml3d Official

Motius Evaluator Humanml3d Official is an evaluator model for text-to-motion matching tasks using the HumanML3D dataset. The tool is provided as a Hugging Face artifact and is described as the official evaluator used by the T2M leaderboard protocol. It includes released inference weights that have been repackaged without retraining, and the optimizer state has been removed. The model is equipped with movement, text, and motion encoders, as indicated by the inclusion of a model.safetensors file. Additional components of the artifact are normalization statistics for HumanML3D-263, word-vector lookup files, configuration files detailing architecture and protocol, and a manifest containing file and source-checkpoint hashes. The evaluator is distributed under the MIT license. Its provenance references a paper on generating diverse and natural 3D human motions from text, and it is based on an original repository named EricGuo5513/text-to-motion. The Motius implementation is identified as HumanML263Evaluator. The tool is intended for use in scenarios where benchmarking or evaluation of text-to-motion models on the HumanML3D dataset is required, such as in leaderboard protocols or research tasks focused on human motion generation from text descriptions. The HumanML3D dataset and its annotations are referenced but remain subject to their original licensing and terms. The model is delivered as a downloadable artifact on Hugging Face, containing all necessary files for evaluation. Further details about compatibility, integration, or usage instructions are not provided in the available evidence.